app.py

from __future__ import annotations

import json
import shutil
import subprocess
import tempfile
from datetime import datetime, timedelta
from functools import lru_cache
from pathlib import Path
from huggingface_hub import hf_hub_download

import gradio as gr

from modular_graph_and_candidates import build_graph_json, generate_html, build_timeline_json, generate_timeline_html, filter_graph_by_threshold

def _escape_srcdoc(text: str) -> str:
    """Escape for inclusion inside an <iframe srcdoc="…"> attribute."""
    return (
        text.replace("&", "&amp;")
            .replace("\"", "&quot;")
            .replace("'", "&#x27;")
            .replace("<", "&lt;")
            .replace(">", "&gt;")
    )


HF_MAIN_REPO = "https://github.com/huggingface/transformers"

CACHE_REPO = "Molbap/hf_cached_embeds_log"

def _fetch_from_cache_repo(kind: str, sim_method: str, threshold: float, multimodal: bool):
    repo_id = CACHE_REPO
    latest_fp = hf_hub_download(repo_id=repo_id, filename="latest.json", repo_type="dataset")
    info = json.loads(Path(latest_fp).read_text(encoding="utf-8"))
    sha = info.get("sha")
    key = f"{sha}/{sim_method}-m{int(multimodal)}"
    json_fp = hf_hub_download(repo_id=repo_id, filename=f"{kind}/{key}.json", repo_type="dataset")

    raw_data = json.loads(Path(json_fp).read_text(encoding="utf-8"))
    filtered_data = filter_graph_by_threshold(raw_data, threshold)

    if kind == "timeline":
        from modular_graph_and_candidates import generate_timeline_html
        raw_html = generate_timeline_html(filtered_data)
    else:
        raw_html = generate_html(filtered_data)

    iframe_html = f'<iframe style="width:100%;height:85vh;border:none;" srcdoc="{_escape_srcdoc(raw_html)}"></iframe>'
    tmp = Path(tempfile.mkstemp(suffix=("_timeline.json" if kind == "timeline" else ".json"))[1])
    tmp.write_text(json.dumps(filtered_data), encoding="utf-8")
    return iframe_html, str(tmp)



def run_loc(sim_method: str, multimodal: bool):
    latest_fp = hf_hub_download(repo_id=CACHE_REPO, filename="latest.json", repo_type="dataset")
    info = json.loads(Path(latest_fp).read_text(encoding="utf-8"))
    sha = info["sha"]
    key = f"{sha}/{sim_method}-m{int(multimodal)}"
    html_fp = hf_hub_download(repo_id=CACHE_REPO, filename=f"loc/{key}.html", repo_type="dataset")
    raw_html = Path(html_fp).read_text(encoding="utf-8")
    iframe_html = f'<iframe style="width:100%;height:85vh;border:none;" srcdoc="{_escape_srcdoc(raw_html)}"></iframe>'
    return iframe_html


def run_graph(repo_url: str, threshold: float, multimodal: bool, sim_method: str):
    return _fetch_from_cache_repo("graph", sim_method, threshold, multimodal)


def run_timeline(repo_url: str, threshold: float, multimodal: bool, sim_method: str):
    return _fetch_from_cache_repo("timeline", sim_method, threshold, multimodal)



# ───────────────────────────── UI ────────────────────────────────────────────────


CUSTOM_CSS = """
#graph_html iframe, #timeline_html iframe {height:85vh !important; width:100% !important; border:none;}
"""

with gr.Blocks(css=CUSTOM_CSS) as demo:
    gr.Markdown("## 🔍 Modular-candidate explorer for 🤗 Transformers")

    # capture the Tabs as a component so we can control selection
    with gr.Tabs() as tabs:
        with gr.Tab("Chronological Timeline", id="timeline"):
            with gr.Row():
                timeline_repo_in = gr.Text(value=HF_MAIN_REPO, label="Repo / fork URL")
                timeline_thresh = gr.Slider(0.50, 0.95, value=0.5, step=0.01, label="Similarity ≥")
                timeline_multi_cb = gr.Checkbox(label="Only multimodal models")
                gr.Markdown("**Embedding method:** TBD")
                timeline_btn = gr.Button("Build timeline")

            timeline_html_out = gr.HTML(elem_id="timeline_html", show_label=False)
            timeline_json_out = gr.File(label="Download timeline.json")

            timeline_btn.click(
                lambda repo, thresh, multi: run_timeline(repo, thresh, multi, "jaccard"),
                [timeline_repo_in, timeline_thresh, timeline_multi_cb],
                [timeline_html_out, timeline_json_out],
            )

        with gr.Tab("LOC Growth", id="loc"):
            sim_radio2 = gr.Radio(["jaccard","embedding"], value="jaccard", label="Similarity metric")
            multi_cb2  = gr.Checkbox(label="Only multimodal models")
            go_loc     = gr.Button("Show LOC growth")
            loc_html   = gr.HTML(show_label=False)
            go_loc.click(run_loc, [sim_radio2, multi_cb2], loc_html)

        with gr.Tab("Dependency Graph", id="graph"):
            with gr.Row():
                repo_in   = gr.Text(value=HF_MAIN_REPO, label="Repo / fork URL")
                thresh    = gr.Slider(0.50, 0.95, value=0.5, step=0.01, label="Similarity ≥")
                multi_cb  = gr.Checkbox(label="Only multimodal models")
                gr.Markdown("**Embedding method:** TBD")
                go_btn    = gr.Button("Build graph")

            graph_html_out  = gr.HTML(elem_id="graph_html", show_label=False)
            graph_json_out  = gr.File(label="Download graph.json")

            go_btn.click(
                lambda repo, thresh, multi: run_graph(repo, thresh, multi, "jaccard"),
                [repo_in, thresh, multi_cb],
                [graph_html_out, graph_json_out],
            )

    def _select_tab_on_load(req: gr.Request):
        tab = (req.query_params or {}).get("tab")
        if tab in {"timeline", "loc", "graph"}:
            return gr.update(selected=tab)
        return gr.update()

    demo.load(_select_tab_on_load, outputs=tabs)

if __name__ == "__main__":
    demo.launch(allowed_paths=["static"])
    demo.launch(allowed_paths=["static"])